Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 716 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Jan Olšan

Senior member
Jan 12, 2017
407
718
136
I am disappointed and surprised at how much AMD stopped giving a dang about INT performance and somehow decided to go all in on what is objectively extreme FP/SIMD preference.
Zen 5 is very interesting and a massive improvement in a lot of special, mostly server/enterprise oriented cases, and kind of a big "BUT WHY?" for almost all standard programs and workloads that you'd expect out of a basic consumer CPU.
Browser benchmarks aren't bad though. Or 1T Geekbench (somewhat under Apple M4, above everything else). Or single-thread performance in apps (as opposed to games). Which should be standard workloads for consumer CPU. Zen 5 just doesn't look that cool due to having the disadvantage of not being able to rely on any clock gains since that lunch was eaten by Zen 4 generation. But the absolute performance is on the top.
 

poke01

Platinum Member
Mar 8, 2022
2,207
2,804
106
Browser benchmarks aren't bad though. Or 1T Geekbench (somewhat under Apple M4, above everything else). Or single-thread performance in apps (as opposed to games). Which should be standard workloads for consumer CPU. Zen 5 just doesn't look that cool due to having the disadvantage of not being able to rely on any clock gains since that lunch was eaten by Zen 4 generation. But the absolute performance is on the top.
the problem with ZEN5 is that’s it’s not balanced. It’s too much server leaning other than that it’s a good core. It’s just not balanced like Zen should be.
 

Jan Olšan

Senior member
Jan 12, 2017
407
718
136
Frankly, I think my only real complaint is that the abuse of FP over INT's general necessity feels like an absurd choice. I can guess why, but it is absurd anyway. It's opening a wide open space for ARM to sit itself pretty in client chips, while server gets courted to an almost obscene degree. We had balance, we now have Google's needs outweighing the general consumer's needs, and I would hope that Z6 fixes that, but...fat chance.
Note that it's more SIMD than floating-point. AVX-512 can work with both floating-point data (HPC usage) and with integer data types - which would be something multimedia code like x264, x265, other encoders, or video and image processing programs would use. Also it's currently used for AI inferencing (VNNI).

Zen 5's AVX-512 splurging targets both floating-point compute and the integer side (some stuff, like very powerful shuffle units will probably be used more on the integer side rather than the floating point side).

Also, traditional (legacy) x87 floating point has not seen much gain, but that was probably not what people even expect these days.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
26,171
15,324
136
the problem with ZEN5 is that’s it’s not balanced. It’s too much server leaning other than that it’s a good core. It’s just not balanced like Zen should be.
I commented elsewhere in this thread. But not only does it have superior effifiency, but having double the avx-512 performance is big for some people. Like me. I think its balanced.
 

del42sa

Member
May 28, 2013
115
153
116
is Zen 5 still good ? Besides AVX 512 and Linux ? ;)
Maybe Zen6 going to be even better and finally we will see regression in fps in games compared to predecessors and maybe even negative IPC. Hope never dies :p
 
Last edited:

yuri69

Senior member
Jul 16, 2013
541
976
136
I commented elsewhere in this thread. But not only does it have superior effifiency, but having double the avx-512 performance is big for some people. Like me. I think its balanced.
Having no AMX is big for some people. Like me. I thing its unbalanced.

Therefore I look forward to Intel server offerings. We have plenty of time till 2027.
 
  • Like
Reactions: del42sa

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
Having no AMX is big for some people. Like me. I thing its unbalanced.

Therefore I look forward to Intel server offerings. We have plenty of time till 2027.
GEMM spam is for accelerators.
I mean sure it still has a niche in a CPU, but you need massive membw to bother, which means HBM.
For a generic DDR CPU, it is overkill.
 
  • Like
Reactions: inquiss

DrMrLordX

Lifer
Apr 27, 2000
22,065
11,693
136
Unironically great uplift in browsers, what the hell.

Also Python, PHP, Node.js interprets are super fast.

Electron. Props to @adroc_thurston for pointing that out ages ago.

Phoronix always runs a huge number and variety of tests.

Yes, and some of the ones they throw at desktop CPUs are hella obscure with peculiar performance profiles. You really have to know your stuff to read a full Phoronix review of a desktop CPU and understand what information is on display. Their server CPU tests are much more concise.

When is Turin expected to launch?

Some probably already have them, through ODMs.

2. Everything else is better served by a product like Sierra Forest

Ahahaha what? Can you buy one? Unless you're on Intel's "do or die" list then you aren't getting one anytime soon. Meanwhile Bergamo filled that niche over a year ago. Sierra Forest only wins on power, which isn't saying much considering how much later it was to market. AMD is already moving on with Turin dense. Clearwater Forest has its work cut out for it!
 

yuri69

Senior member
Jul 16, 2013
541
976
136
He's not the only one that's said that's what's on AMD's roadmap...I'm afraid it might be true.

Very much hoping it's not and I have to agree with the others that said that AMD may as well give up if they take that long.

Counter point to that: Everyone was wrong about RDNA3, and that was downstream of bad information coming out of AMD. Being wrong (for the right reasons) about RDNA3 while being right about Navi4c's cancellation is not incongruous.

Being 180 degrees wrong about Zen 5 IPC, while also somehow being right about Zen 6 intentionally being targeted so late would be incongruous. Because if you have a robust source and are not engaging in inappropriate extrapolation, you shouldn't be getting only one of those right (again, for the right reasons. You could easily end up right on Zen 6 for the wrong reasons).

And if you know anything about the way management tends to set goals, or how engineers tend to make predications, and it's odd to think that you wouldn't, then 2027 Zen 6 stuff feels so out there that even if you do have a robust source this time around, even if it's coming out of AMD itself, rather than say, reading the tea leaves on some very specific server part based from some far downstream third party source, then, if I were you I'd still be extremely cautious that AMD isn't intentionally putting out bad information as part of an infosec strategy.

Like, the only way it seems it could possibly be right is if AMD has some stuff coming down the pipe in the meantime it feels is competitive, while Zen 6 is an extreme outlier in terms of performance increase. Otherwise, it's like, just about everything seems like it's theoretically doable in a two year span. So planning for the failure with a 'tick' that only comes out three years after an already significantly delayed 'tock', with Zen5c already acting as a 3nm pipecleaner... it simply strains credulity.

You can say "it aligns with DDR6," and that sounds not totally insane until you think about when Zen 5 would have been planned to come out. At that point we're be talking about an intentional four year gap. Was AMD assuming that Intel would implode in the meantime? Insanity.
Clowns who hyped RDNA3 were simply fooled by the 2x ALUs. People counter argumenting those fantasies with "that 128b chip has no bandwidth to be 2x" or "see LLVM patches of the super-limited double-issue" were trashed.

The Zen 6 in 2027 is odd but doable:

CONS:
* AMD must supply OEMs with mobile chips. For 2025 - that's Bald Eagle Point - but also they need something for 2026 - a refresh of refresh as the primary chip would be... weird.
* 24+ months is AMD core IP cadence but this would be way longer

PROS:
* AMD struggles to handle their now-large IP/SKU portfolio - even with current budgets they can't put their AGESA together.
* The core IP cadence gets longer and longer since Zen 3-4, so it is likely to continue the trend.
* Dependency on new tech - Zen 6 uses a completely new packaging technology & maybe even a new RAM standard or TSMC node - all deps need to be *cheaply* available for Zen 6.
* Zen 5 was probably very expensive to develop - they need to handle the ROI.
 

coercitiv

Diamond Member
Jan 24, 2014
6,678
14,275
136
Also Python, PHP, Node.js interprets are super fast.
Browser performance + all of the above is the main reason why Zen 5 is still on the table for me, though on the home desktop I'd have to go 3D to make my purchase worthwhile. Not going to rush at all though, will wait for Arrow Lake launch, maybe even more to let prices adjust to demand.

2024 was supposed to be a great year for CPU upgrades with so many new products, but seems like all of them are shaping up to be mediocre in some way or the other.
 

del42sa

Member
May 28, 2013
115
153
116
Kind of irony thing is that for ZEN5 they had the biggest budget to develop a new ground up architecture redesing (the same RDNA3 ) and yet the result is very unbalanced and embarrassing. It seems that since the release of the first Zen, they have been much better at this and with kind of minor changes achieved much better results ( Zen3/ RDNA2 ) Perhaps it´s just a phase and next CPU arch will make things more straight
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
List of all the things leakers got wrong
If I'm honest, almost all speculation is a mix of raw invention, hearsay, wishful thinking and personal bias from limited technical understanding. Plus, you know, corpos not wanting their secrets spilled.
It's one of the reasons I kind of disappeared, the whole speculation lifestyle is definitely not going anywhere except in circles.
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
Kind of irony thing is that for ZEN5 they had the biggest budget to develop a new ground up architecture redesing (the same RDNA3 ) and yet the result is very unbalanced and embarrassing. It seems that since the release of the first Zen, they have been much better at this and with kind of minor changes achieved much better results ( Zen3/ RDNA2 ) Perhaps it´s just a phase and next CPU arch will make things more straight
In a somewhat better light, you could say that Zen 2 was also very heavily productivity oriented and poor at games and such. It's not impossible that they went for that first and will re-balance later. But yes, if it's going to be 18 months to two years of Zen 5 like it was with Zen 4 (or worse?), it's definitely not a super fun prospect to have noisy arguments about "no improvements since 2022 in gaming".
 

coercitiv

Diamond Member
Jan 24, 2014
6,678
14,275
136
I believe a degree of anger went through his fingers there.
Also, AMD's marketing is truly abysmal.
They are no longer to be trusted. Like I said before, they wasted trust capital built in many years, for both CPUs and GPUs.

No claim from AMD is to be trusted anymore. It's even worse if you consider the lies for the recently launched AM4 SKUs, those did not present high stakes for AMD and the marketing team still lied through their teeth.
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
Yes, AMD was heavily trustworthy and generally doubt was cast much more heavily on Intel/Nvidia...but this isn't a "big lie" like RDNA 3. This is more like the second or third big lie in a row. Zen 5 isn't what was apparently advertised to our leakers. Nor the 5900 XT. Nor RDNA 3. Ironically, RDNA 3.5, which fixes a lot of RDNA 3, and will be their best product till RDNA 4, is also somehow glossed over without any performance statement.

God their marketing is awful, replace it with an angry vomiting poisonous frog already, at least it's an improvement.
 

poke01

Platinum Member
Mar 8, 2022
2,207
2,804
106
God their marketing is awful, replace it with an angry vomiting poisonous frog already, at least it's an improvement.
They should do fun graphs like Apple does, they might be crap but at least it’s honest about being crap upfront lol.

Wouldn’t make any difference at this point.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,539
3,471
106
Looking like a long reign for the 7800X3D gaming champ.

Other tests show web browser performance improvement. I guess they did figure out a way to target improved interpreter and JIT performance.

What would be the argument for 9800x3d not getting at least equal performance bump (from 9700x) as 7800x3d (from 7700x)?

It seems to me that 9800x3d will receive at least equal performance bump, and some things (lower power, lower temps) suggest that it is more likely that 9800x3d will not have the same clock speed regression as 7800x3d.
 

FlameTail

Diamond Member
Dec 15, 2021
3,952
2,376
106
What would be the argument for 9800x3d not getting at least equal performance bump (from 9700x) as 7800x3d (from 7700x)?

It seems to me that 9800x3d will receive at least equal performance bump, and some things (lower power, lower temps) suggest that it is more likely that 9800x3d will not have the same clock speed regression as 7800x3d.
There were rumours that V-cache in Ryzen 9000X3D is different.
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
I'm not expecting anything fancy, already 2nd gen Vcache was a lot better than 1st gen. It's probably a few extra percent, 2-3% tops. Already the amount of cycles lost by using V cache was tiny, I think someone mentioned just 3-4 cycles. There's just not a massive headroom there IMO.

In any case, this is it then. A 70mm², decent density, power efficiency oriented, productivity monster Zen 5 with fairly crappy gaming performance. That's the end result of our forever waiting.
Well, it's both good and completely meh. Server lads will absolutely buy it in infinity quantity, and I'm really eager to see if the microcode updates can somehow get better results out of their relatively poor performance.

The best thing about Zen 5 will be Turin. A true cloud chip.
 

yottabit

Golden Member
Jun 5, 2008
1,496
529
146
What would be the argument for 9800x3d not getting at least equal performance bump (from 9700x) as 7800x3d (from 7700x)?

It seems to me that 9800x3d will receive at least equal performance bump, and some things (lower power, lower temps) suggest that it is more likely that 9800x3d will not have the same clock speed regression as 7800x3d.
At this point my biggest worry is that AMD already hyped up Zen5 vcache a bit saying it had some “interesting” features

Based on what we’ve seen so far of the core I’d expect it to have even more uplift than Zen4X3D vs Zen4, but if we apply the “reverse marketing lingo” filter, “interesting” doesn’t sound too good….